Weakly supervised vessel segmentation in X-ray angiograms by self-paced learning from noisy labels with suggestive annotation

NEUROCOMPUTING

卷:417

Article

Zhang, Jingyang^1,4  Wang, Guotai²  Xie, Hongzhi³  Zhang, Shuyang³  Huang, Ning⁵  Zhang, Shaoting⁵  Gu, Lixu^1,4 

[1] Shanghai Jiao Tong Univ, Sch Biomed Engn, Shanghai, Peoples R China

[2] Univ Elect Sci & Technol China, Sch Mech & Elect Engn, Chengdu, Peoples R China

[3] Chinese Acad Med Sci, Peking Union Med Coll, Peking Union Med Coll Hosp, Dept Cardiol, Beijing, Peoples R China

[4] Shanghai Jiao Tong Univ, Inst Med Robot, Shanghai, Peoples R China

[5] SenseTime Res, Shanghai, Peoples R China

关键词: Convolutional neural network;    Weakly supervised learning;    Self-paced learning;    Suggestive annotation;    Vessel segmentation;   

DOI  :  10.1016/j.neucom.2020.06.122

来源: Elsevier

【 摘 要 】

The segmentation of coronary arteries in X-ray angiograms by convolutional neural networks (CNNs) is promising yet limited by the requirement of precisely annotating all pixels in a large number of training images, which is extremely labor-intensive especially for complex coronary trees. To alleviate the burden on the annotator, we propose a novel weakly supervised training framework that learns from noisy pseudo labels generated from automatic vessel enhancement, rather than accurate labels obtained by fully manual annotation. A typical self-paced learning scheme is used to make the training process robust against label noise while challenged by the systematic biases in pseudo labels, thus leading to the decreased performance of CNNs at test time. To solve this problem, we propose an annotation-refining self-paced learning framework (AR-SPL) to correct the potential errors using suggestive annotation. An elaborate model-vesselness uncertainty estimation is also proposed to enable the minimal annotation cost for suggestive annotation, based on not only the CNNs in training but also the geometric features of coronary arteries derived directly from raw data. Experiments show that our proposed framework achieves 1) comparable accuracy to fully supervised learning, which also significantly outperforms other weakly supervised learning frameworks; 2) largely reduced annotation cost, i.e., 75.18% of annotation time is saved, and only 3.46% of image regions are required to be annotated; and 3) an efficient intervention process, leading to superior performance with even fewer manual interactions. (C) 2020 Elsevier B.V. All rights reserved.

【 授权许可】

Free